You are on page 1of 11

Progressive Mesh Densification

Method for Numerical Solution


Wei Pu of Mixed Elastohydrodynamic
School of Aeronautics and Astronautics,
Sichuan University,
Chengdu 610065, China
Lubrication
Numerical solution of mixed elastohydrodynamic lubrication (EHL) is of great impor-
Jiaxu Wang1 tance for the study of lubrication formation and breakdown, as well as surface failures of
School of Aeronautics and Astronautics,
mechanical components. However, converged and accurate numerical solutions become
Sichuan University,
more difficult, and solution process with a fixed single discretization mesh for the solution
Chengdu 610065, China;
domain appears to be quite slow, especially when the lubricant films and surface contacts
State Key Laboratory of
coexist with real-machined roughness involved. Also, the effect of computational mesh
Mechanical Transmission,
density is found to be more significant if the average film thickness is small. In the present
Chongqing University,
study, a set of sample cases with and without machined surface roughness are analyzed
Chongqing 40044, China
through the progressive mesh densification (PMD) method, and the obtained results are
e-mail: cquwjx@foxmail.com
compared with those from the direct iteration method with a single fixed mesh. Besides,
more numerical analyses with and without surface roughness in a wide range of operat-
Dong Zhu ing conditions are conducted to investigate the influence of different compound modes in
School of Aeronautics and Astronautics, order to optimize the PMD procedure. In addition, different initial conditions are used to
Sichuan University, study the effect of initial value on the behaviors of this transient solution. It is observed
Chengdu 610065, China that, no matter with or without surface roughness considered, the PMD method is stable
for transient mixed EHL problems and capable of significantly accelerating the EHL
solution process while ensuring numerical accuracy. [DOI: 10.1115/1.4031495]

Keywords: elastohydrodynamic lubrication (EHL), mixed EHL, mixed lubrication,


numerical solution, PMD method, mesh density

1 Introduction mesh is constantly changed with mesh density increasing and


decreasing alternately. Through the repeated transitions between
Power and motion are transmitted through surface contacts in
coarse and fine meshes in either a V- or W-cycle, the total error is
mechanical components, most of which are lubricated. Counter-
fast minimized. Venner [9], Kim and Sadeghi [10], Ai [11], Zhu
formal contacts widely exist in many machine elements, such as
and Ai [12], and many other researchers followed and improved
various types of gears, rolling element bearings, cam–follower
this approach. As a result, solution convergence has been consid-
systems, and others. Thus, EHL is of great importance, especially,
erably accelerated and numerical accuracy was improved. More
thin-film and mixed EHL are the major modes of lubrication in
detailed descriptions for the MG method can be found in Refs.
which most counterformal contact components operate. They also
[8,9] and a book by Venner and Lubrecht [13].
represent critical transition toward lubrication breakdown and sur-
In the last 20 years, the great advancement of computer technol-
face failures. Therefore, a good understanding of thin-film
ogies has fueled significant development in the mixed EHL theory
and mixed EHL characteristics is vital to component design opti-
and practice. Basically, the EHL equation system exhibits very
mization, durability/reliability improvement, and power loss
strong nonlinear behaviors, resulted mainly from the surface elas-
minimization.
tic deformation and much increased viscosity due to pressure.
Early studies for point contact EHL with smooth surfaces under
Therefore, numerical solution under practical conditions has long
the full-film lubrication condition include those by Ranger et al.
been a great challenge to researchers, and convergence difficulty
[1], Hamrock and Dowson [2–4], Evans and Snidle [5,6], Zhu and
often occurs especially when the load is heavy and/or the velocity
Wen [7], Lubrecht [8], Venner [9], Kim and Sadeghi [10], Ai
is low resulting in a very thin or zero lubricant film. Great efforts
[11], and others. Due to limited computational ability in the 1950s
have been made in order to solve this convergence problem, and
through early 1980s, relatively coarse discretization meshes were
one of the developed numerical algorithms is the so-called
employed for solving the Reynolds equation. In each case, at that
“semisystem approach” presented by Ai [11] and improved by
time, a single mesh for the solution domain was used and always
Zhu and Hu [14], Hu and Zhu [15], and others. With this
kept unchanged throughout the entire solution process (see Refs.
approach, numerical simulation of the entire transition of lubrica-
[1–7]). Lubrecht [8] was the first one who introduced the multi-
tion status from the full-film and mixed EHL down to dry contact,
grid (MG) method into EHL solution procedure and significantly
using a unified model with machined roughness, has become pos-
increased the mesh density. The basic idea of the MG method is
sible, and numerical convergence can be ensured even for thin-
that numerical errors whose wavelengths are of the order of the
film and zero-film cases under severe operating conditions.
mesh spacing are fast to reduce and hence can vanish quickly,
During the work reported in Refs. [14,15] and others, it was
implying that high-frequency errors can be reduced quickly on a
found through numerical trials that EHL solution convergence is
finer mesh while low-frequency errors can be removed efficiently
much easier to achieve under severe conditions by turning off the
on a coarser mesh. Thus, in an MG procedure, the computational
MG function and employing the direct iteration method with a
1
fixed single mesh. Repeatedly changing the mesh density up and
Corresponding author.
Contributed by the Tribology Division of ASME for publication in the JOURNAL
down in the MG procedure may not help with the solution conver-
OF TRIBOLOGY. Manuscript received February 8, 2015; final manuscript received gence when the EHL film is very thin. The reason was not well
August 13, 2015; published online October 15, 2015. Assoc. Editor: Zhong Min Jin. understood at that time. Morales-Espejel et al. [16] were the first

Journal of Tribology Copyright V


C 2016 by ASME APRIL 2016, Vol. 138 / 021502-1

Downloaded From: http://tribology.asmedigitalcollection.asme.org/ on 05/06/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use


to point out that the effect of mesh density on the EHL solution the proposed mesh level of highest density is obtained with a
becomes much more significant when the EHL film is very thin. tightened convergence criterion. In this way, the errors with larger
This observation was soon confirmed by Zhu [17], Venner [18], wavelengths are removed first, and then those with intermediate
and Liu et al. [19]. Based on this, Zhu [20] conducted a systematic and small wavelengths are canceled step by step as the mesh con-
study on the mesh density effect in a wide range of operating con- tinues to be progressively densified. Since the computation is very
ditions. It has been revealed that the mesh density effect may be fast on the lower levels of mesh density, the total time required
insignificant when the EHL film is thick, but becomes significant for the ultimate solution can be greatly shortened, as demonstrated
if the film thickness falls down to the order of a couple of tens of by the preliminary results from a few sample cases of smooth sur-
nanometers or less, where the EHL solution depends considerably face EHL in circular contacts, presented in Ref. [20].
on the density of computational mesh used. Using different In the present study, the PMD method is further explored based
meshes, in other words, would aim at (and eventually yield) dif- on the preliminary study in Ref. [20] by analyzing a large number
ferent solutions. It has also been pointed out in Ref. [20] that, with of mixed EHL cases with and without real-machined roughness
the second-order backward differential scheme for the Couette involved. The obtained results are compared with those from the
flow term, commonly used by many researchers, the denser the direct iteration method with a single fixed computational mesh.
mesh, the greater the central film thickness is obtained. In other Besides, the effect of different compound modes on the EHL solu-
words, if HK is defined as the converged dimensionless central tion process is systematically investigated in a wide range of oper-
film thickness on mesh density level K, one can always find ating conditions in order to optimize the PMD procedure. In
HI < HII < HIII <       < HK < HKþ1…. However, when the mesh addition, different initial conditions are used to study the effect of
density is continuously increasing, the film thickness increase is in initial value on the transient solution for the purpose of examining
fact limited. Actually, it quickly approaches an asymptotic limit, the stability of this solution method. It is expected that, no matter
H1, which is defined as the dimensionless film thickness theoreti- with or without surface roughness, the PMD method is stable for
cally from an infinitely dense mesh. Note that H1 can be readily this transient problem and capable of significantly accelerating the
estimated by a simple formula given in Ref. [20]. mixed EHL solution process.
The above-mentioned studies suggest that the MG method may
work well when the EHL film is thick, but it may not be desirable 2 Basic Equations
in case of thin-film and mixed EHL. The reason is obvious: Due
to the significant effect of mesh density on the EHL solution, each The basic formulation for the deterministic mixed EHL simula-
time when the mesh changes, say from level K to level K þ 1, the tion model and its improvements has been presented in a good
solution process changes its target from HK to HKþ1 (which is dif- number of previous publications, such as Refs. [14,15,19,20]. Its
ferent from HK), hence repeatedly changing the mesh in a W- or details will not be repeated here but a brief summary is given in
V-cycle would make the solution process aiming at different tar- the following part of Sec. 2 for clarity and completeness. The
gets that oscillate up and down repeatedly. This would possibly present study focuses on the mixed EHL problems in counterfor-
jeopardize the process of achieving the converged ultimate solu- mal contacts, for which the Hertzian contact zone can be either an
tion especially when the EHL film is very thin. There are a few infinitely long narrow band (line contact) or an ellipse (elliptical
other reasons for the MG method not being preferred for the thin- contact). Generally, two arbitrary rough surfaces are running
film and mixed EHL, especially if high-frequency machined through the EHL conjunction at any rolling and relative sliding
roughness is involved. Detailed discussions can be found in velocities. The two rough surfaces can be directly digitized and
Ref. [20]. then used as a part of the input data. The pressure (including both
Although solution convergence has been well ensured, the com- the “hydrodynamic” pressure and “contact” pressure) within the
putational efficiency is still a major concern during an EHL solu- entire domain is governed by the Reynolds equation expressed as
tion process. In practice, a high-density mesh is preferred for thin- follows:
film and mixed EHL in order to achieve satisfactory numerical ! !
accuracy, especially when high-frequency machined roughness is @ qh3 @p @ qh3 @p u1 þ u2 @ ðqhÞ @ ðqhÞ
involved. However, it is well known that, for any computational þ ¼ þ (1)
@x 12g @x @y 12g @y 2 @x @t
analysis based on numerical discretization, the higher the mesh
density, the smaller the discretization error but the longer the Here, the lubricant can be either Newtonian or non-Newtonian (If
computing time. Generally, when mesh density is doubled in both non-Newtonian, g should be the effective viscosity with respect to
the x- and y-directions, the computational time required for each a certain shear behavior of the lubricant used. Readers may refer
iteration is increased approximately by a factor of four. In the to Refs. [14,15] for more information). All the parameters are
meantime, the change of converged EHL film thickness solution defined in the Nomenclature. The x-coordinate is chosen to coin-
between two subsequent meshes reduces roughly by the same fac- cide with the rolling direction. A boundary condition of pressure
tor of four, if a second-order discretization scheme is used and the p ¼ 0 at the edges of the solution domain, as well as a normal cav-
mesh densification continues. Based on the necessary compromise itation condition, must be satisfied when solving this equation.
between the numerical accuracy and the computational efficiency, For a rough surface point contact EHL problem, the local lubri-
the discretization mesh can be optimized, and the range of reason- cant film thickness/gap is time-dependent and can be computed by
able mesh density has been suggested in Ref. [20].
When a reasonably dense fixed single mesh is employed for x2 y2
ensuring the solution convergence and numerical accuracy, errors h ¼ h0 ðtÞ þ þ þ d1 ð x; y; tÞ þ d2 ð x; y; tÞ þ V ð x; y; tÞ (2)
2Rx 2Ry
whose wavelengths are comparable to the mesh spacing can be
reduced quickly, but low-frequency errors with considerably Note that the first term on the right-hand side of Eq. (2) represents
larger wavelengths may take a very long time to vanish. In order the normal approach between the two bodies, and the second and
to accelerate the solution process, a PMD method was initially third terms are for the original macrocontact geometry. V(x, y, t) is
proposed by Zhu [20]. With this approach, solution process is the surface elastic deformation, and d1(x, y, t) and d2(x, y, t) are
started on a low level of mesh density, so that numerical errors the original roughness profiles for surfaces No. 1 and No. 2,
with relatively long wavelengths can be removed and the first respectively. Since both surfaces are moving along the
approximation of the solution can be obtained very quickly. As x-direction, d1 and d2, are time-dependent, and can be given by
soon as the solution becomes reasonably convergent with a rela-
tively loose convergent criterion, the computation is automatically
switched to the next level for a higher (typically doubled) mesh d1 ðx; y; tÞ ¼ s1 ðx  u1 t; yÞ
density. This procedure is repeated until the ultimate solution on d2 ðx; y; tÞ ¼ s2 ðx  u2 t; yÞ

021502-2 / Vol. 138, APRIL 2016 Transactions of the ASME

Downloaded From: http://tribology.asmedigitalcollection.asme.org/ on 05/06/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use


where s1(x, y) and s2(x, y) are surface profiles, typically in the 3 PMD Method
form of measured discretized roughness height data matrix.
The theoretical basis of the PMD method is similar to that of
The surface elastic deformation, V(x, y, t), can be calculated by
the MG procedure, i.e., errors whose wavelengths are of the order
the Boussinesq integral written as
of the mesh size can be quickly reduced by using an iterative
ðð relaxation procedure, while other error components with the
2 pðn; 1Þ
V ð x; y; tÞ ¼ 0 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi dnd1 (3) wavelengths far away from the mesh spacing may take a much
pE ð x  n Þ 2
þ y  1 2 longer time to vanish. In fact, if the wavelength of the error is not
X ð Þ
comparable to the mesh size, it may often hardly reduce and, con-
The lubricant viscosity is assumed to be dependent on pressure, sequently, after a few iterations convergence slows down and the
and one of the commonly used viscosity equations is the exponen- asymptotic convergence rate may become very small (see Refs.
tial law given below [9,13]). That is why different mesh levels may be needed in order
to accelerate the entire solution process. However, the major dif-
g ¼ g0 eaP (4) ference between the PMD and the MG is that with the MG the
mesh density is going up and down repeatedly in a V- or W-cycle,
Other viscosity laws can also be used here to readily replace while in the PMD procedure the computation is started from a
Eq. (4) when necessary. Detailed discussion on the effect of dif- low-density mesh, then the mesh is progressively densified step
ferent viscosity models is beyond the scope of the present study. by step until the proposed highest mesh density is reached. When
The density of lubricant is also a function of pressure, commonly the unidirectional mesh densification process continues, the low-
calculated by frequency errors are canceled first, then those of the intermediate
! and high frequencies. During this process, the first approximation
0:6  109 p of the solution is achieved very quickly then the solution gradu-
q ¼ qo 1 þ (5)
1 þ 1:7  109 p ally approaches its converged ultimate target unidirectionally.
Since the majority of iterations are on the low-level meshes, the
The applied load is balanced by the integral of the pressure entire solution process can be very fast.
over the entire solution domain, X, i.e., Figure 1 shows an example for different levels of mesh density,
in which level I mesh consists of 32  32 grids for the entire
ðð domain, level II is 64  64, level III 128  128, and so on. If K is
wðtÞ ¼ pðx; y; tÞdxdy (6) defined as the current mesh level number, in the beginning of a
X PMD procedure, a low-level mesh, usually K ¼ II or III, is
employed in order to quickly cancel low-frequency errors and get
The numerical solution techniques based on the semisystem good approximations of the pressure and film thickness distribu-
approach have been described previously in Refs. tions. Due to the low mesh density, the computation is very fast.
[11,14,15,19,20] and some other publications. The unified As soon as a relatively loose convergence criterion is met, such as
4
lubrication-contact approach employs the same equation system ep ¼ RjPnew old new
i;j  Pi;j j=RPi;j < 10 , the computation is automati-
and numerical algorithms consistently in both the hydrodynamic cally switched to the next higher level of mesh to focus on the
and contact regions. In order to accelerate the surface deformation reduction of errors of higher frequencies, pursuing a more accu-
calculation, which is a major part of the total computation, the dis- rate solution. This procedure is repeated until the proposed ulti-
crete convolution and fast Fourier transform method developed by mate mesh density is reached. On the final mesh level of the
Liu et al. [21] have been incorporated into the EHL model men- highest density, the convergence criteria may be tightened, such
tioned above by Wang et al. [22] and others, and the computation as ep < 106 , to ensure better solution accuracy.
using this set of numerical procedure has been proven to be very Each time when the mesh is changed from a low level to a
fast. When solving mixed EHL problems with machined rough- higher level, the nodal pressures and film thickness values are
ness, the handling of discretization meshes and the resulting transferred through interpolation. In the present study, the second-
numerical accuracy become crucial. The appropriate algorithm order Lagrangian (parabolic) interpolation is used and found to be
has been described in detail in a recent paper by Zhu et al. [23]. satisfactory.

Fig. 1 Different mesh density levels

Journal of Tribology APRIL 2016, Vol. 138 / 021502-3

Downloaded From: http://tribology.asmedigitalcollection.asme.org/ on 05/06/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use


Table 1 Different mesh levels used DX ¼ DY ¼ 0.005859, which is sufficiently small. Generally, a
small time step length is preferred for transient mixed EHL prob-
Mesh level, K Mesh density, DM Mesh size, DX ¼ DY lems in order to obtain stabilized numerical solutions. However, a
small step length may cause a considerably increased total com-
I (32  32) 10.667 0.09375 puting time. According to Lubrecht and Venner [24] and Mourier
II (64  64) 21.333 0.046875
III (128  128) 42.667 0.0234375
et al. [25], a dimensionless time step length equal to or smaller
IV (256  256) 85.333 0.01171875 than the dimensionless mesh size in the x-direction is appropriate,
V (512  512) 170.67 0.005859375 which is in agreement with Zhu and Hu’s observation presented in
Ref. [26]. In the present study, therefore, the time step length is
chosen to be 0.0044, slightly smaller than the mesh size at the
highest level. The pressure convergence criterion is set to be
ep < 106 for smooth surface solutions and ep < 104 if machined
roughness is involved. The mesh density, defined as DM ¼ 1/
DX ¼ 1/DY, at each level can be found in Table 1.
It is well known that converged and accurate numerical solu-
tions for thin-film and mixed EHL become more difficult than
those in thick-film cases, and the solution process with a fixed sin-
gle mesh appears to be quite slow. Since the problem for mixed
EHL is usually transient due to moving surface irregularities, it
may take many time steps to reach a globally stabilized solution.
In this section, some sample cases are numerically analyzed under
moderate and low-speed conditions with and without machined
surface roughness considered. When machined roughness is
involved, two surfaces from the same grinding process are in con-
tact against each other, and a sample ground rough surface topog-
Fig. 2 A sample three-dimensional ground surface profile raphy can be found in Fig. 2. Four typical cases with and without
surface roughness at the rolling speed of U ¼ 1 m/s and
4 Comparison Between PMD and That With Single U ¼ 0.01 m/s are shown in Figs. 3 and 4, demonstrating the com-
parison between the PMD method and that with a single fixed
Fixed Mesh computational mesh of K ¼ V. In this set of cases, the initial value
Sample cases in a circular contact (a ¼ b) are analyzed in the of pressure distribution is from the Hertzian dry contact, the
present study with a solution domain set to be 1:9  X  1:1 applied load is fixed at 1000 N, and the material properties
and 1:5  Y  1:5. The corresponding dimensionless mesh are go ¼ 0.09 Pa  s and E0 ¼ 230.77 GPa. The pressure–viscosity
spacing at the highest level, K ¼ V (512  512 grids), is coefficient is 12.5 GPa1. The geometric parameters are

Fig. 3 Comparison of smooth surface solutions between PMD method and that with a fixed
mesh of K 5 V: (a) rolling speed of U 5 1 m/s and (b) rolling speed of U 5 0.01 m/s

021502-4 / Vol. 138, APRIL 2016 Transactions of the ASME

Downloaded From: http://tribology.asmedigitalcollection.asme.org/ on 05/06/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use


Fig. 4 Comparison of rough surface solutions between PMD method and that with a fixed
mesh of K 5 V: (a) rolling speed of U 5 1 m/s and (b) rolling speed of U 5 0.01 m/s

Rx ¼ Ry ¼ 20 mm, so that the Hertzian contact radius is PMD, on the other hand, it takes only 150 iterations or so for
a ¼ b ¼ 0.5066 mm and the maximum Hertzian pressure U ¼ 1 m/s and about 175 iterations for U ¼ 0.01 m/s on the low-
Ph ¼ 1.86 GPa. For the rough surface cases, the composite rms level meshes to get good approximations of the solution. On the
roughness of the two transversely oriented ground surfaces is set highest mesh level of K ¼ V, only a small number of iterations,
to be r ¼ 0.50 lm. say less than 50, are needed by the PMD. Therefore, the total
Note that in Fig. 4, H is the dimensionless average film thick- computing time is greatly reduced.
ness (or gap) calculated in the central part of Hertzian contact (see For the cases shown in Fig. 4, both the surfaces are rough with
Refs. [14,15] for the definition of average film thickness). machined topography. In these cases, the PMD method is kept the
In the cases shown in Fig. 3, both the surfaces are smooth. In same as those given above for the cases shown in Fig. 3. It should
Fig. 3(a), the rolling speed is U ¼ 1 m/s and in Fig. 3(b) be noted that, in the present study, for all the transient cases, the
U ¼ 0.01 m/s. This two figures show the variation of dimension- number of iteration in each figure represents total number of time
less central film thickness versus the number of iteration, and step. Obtained ultimate average film thickness are 146.9 nm and
direct comparison between PMD method and that with a fixed 63.4 nm, and contact load ratio, Wc, are 57.12% and 84.98%, for
mesh of K ¼ V is made under otherwise the same conditions. With U ¼ 1 and 0.01 m/s, respectively, with a reasonably good conver-
the PMD method, the analysis is started at K ¼ II and then gence accuracy of ep < 104 and eW < 104. When the computa-
switched to K ¼ III, K ¼ IV, and eventually to K ¼ V. Obtained tion stays on the mesh level K ¼ V all the time from the
ultimate central film thicknesses are 338.78 nm and 10.98 nm beginning, the CPU time required to get comparable solutions
for U ¼ 1 m/s and U ¼ 0.01 m/s, respectively, with a high (146 nm for U ¼ 1 m/s and 63.4 nm for U ¼ 0.01 m/s) will be about
convergence accuracy of ep < 106 and eW ¼ jW new  W old j 1.87 and 2.24 times of that by the PMD approach, respectively.
=W new < 106 . When the single fixed mesh of K ¼ V is used, i.e., Therefore, the PMD method appears to be able to significantly
computation stays on the high-density mesh level of K ¼ V all the improve the computational efficiency.
time from the beginning, the central processing unit (CPU) time It is important to note that, regardless of the rolling speed and
required to get comparable solutions (335.49 for U ¼ 1 m/s and the roughness, after a sufficient number of time steps (such as
11.15 nm for U ¼ 0.01 m/s) will be about 26 and 14.9 times of that about 1400 time steps in Fig. 4(b), equivalent to a dimensionless
by the PMD, respectively, if the numerical algorithms are other- time of 6.16), the obtained solutions from the PMD and the direct
wise exactly the same. It is observed that, when the direct iterative iteration with a fixed single mesh will be about the same and the
method with a single fixed dense mesh is employed, it takes many discrepancy will be negligible. This means that the final mixed
iterations (and a very long time) to get the solution roughly shaped EHL solution is in principle independent of the intermediate mesh
due to very slow reduction of the low-frequency errors. With the arrangements, and the solution method is stable and reliable.

Journal of Tribology APRIL 2016, Vol. 138 / 021502-5

Downloaded From: http://tribology.asmedigitalcollection.asme.org/ on 05/06/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use


Table 2 Different compound modes 5 Different Compound Modes for PMD Procedure
Scheme Compound mode description In the above study with the PMD method, the analyses are
started from mesh level K ¼ II, and then progressively switched to
Scheme 1 From level II to level III and then jump to level V K ¼ III, IV, and eventually to K ¼ V. This compound mode may
Scheme 2 From level II jump to level IV and then to level V not be the best, if some errors of certain intermediate wavelengths
Scheme 3 From level II directly jump to level V are insignificant thus iterations on one mesh level or even two
Scheme 4 From level II to III and IV then to level V may be skipped in order to further increase the speed of solution
process. Since the frequency spectrum of numerical errors is
unknown, in this section we will optimize the PMD procedure by
conducting a good number of trials employing different com-
pound modes with and without surface roughness under different
operating conditions. Details of the compound modes used are
Table 3 Results of central film thickness with smooth surfaces listed in Table 2.
Four different entraining velocities are selected for the numeri-
Converged central film thickness (nm)
cal trials: 10 m/s, 1 m/s, 0.1 m/s, and 0.01 m/s, resulting in EHL
Case no. U (m/s) HII HIII HIV HV film thickness variation in a wide range so that the present investi-
gation is representative. For the trials with smooth surfaces, the
1 10 1624.83 1657.81 1666.84 1669.26 range of central film thickness is from 1669.26 nm down to
2 1 287.84 332.06 343.89 347.28 11.15 nm, if all the other input parameters remain the same as
3 0.1 0 46.54 62.03 66.30 those given above. Obtained main results are summarized in Table
4 0.01 0 0 5.92 11.15 3. For those with the rough surfaces, the obtained numerical solu-
tions correspond to a range of average film thickness from

Fig. 5 Comparison of results from different schemes with smooth surfaces: (a) rolling speed of U 5 1 m/s and (b) rolling speed of
U 5 0.01 m/s

021502-6 / Vol. 138, APRIL 2016 Transactions of the ASME

Downloaded From: http://tribology.asmedigitalcollection.asme.org/ on 05/06/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use


Fig. 5 Continued

1136.8 nm down to 63.4 nm, and the contact load ratio varies from
zero to 84.98%, which represents a transition of lubrication status
from the full-film and mixed EHL down to boundary lubrication
(or nearly dry contact).
The more detailed results at U ¼ 1 m/s and U ¼ 0.01 m/s are
illustrated in Fig. 5 that shows the iteration process for different
schemes with smooth surfaces. For each scheme, the final film
thickness contour on each mesh level is given, and the variation of
dimensionless central film thickness versus the number of itera-
tion is plotted. Note that, no matter which compound mode is
used, the higher the mesh density, the greater the converged film
thickness, i.e., HII < HIII < HIV < HV. Also, the mesh density
effect is certainly more significant in the thin-film case at
U ¼ 0.01 m/s than that in the thick-film case at U ¼ 1 m/s, as
shown in Fig. 5. This is in good agreement with the observations
reported in Refs. [19,20].
The comparison of required computing time with different
schemes is given in Fig. 6. Note that in the figure Ti (i ¼ 1, 2, 3, 4)
is defined as the computing time required by scheme i, while Td is
that by the direct iteration method with the fixed mesh of K ¼ V.
Fig. 6 Comparison of computational speeds with different It is important to note that the total computing time may not have
schemes any direct relation with the total number of iterations. For

Journal of Tribology APRIL 2016, Vol. 138 / 021502-7

Downloaded From: http://tribology.asmedigitalcollection.asme.org/ on 05/06/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use


Fig. 7 Comparison of different schemes with rough surfaces: (a) solution details at U 5 1 m/s, (b) summary for solutions at
U 5 1 m/s, (c) solution details at U 5 0.01 m/s, and (d) summary for solutions at U 5 0.01 m/s

example, if most iterations are conducted on the coarse meshes, 0.01 m/s or lower, the computational speed by scheme 4 becomes
the total computing time can be short even though the total num- slightly lower than that by scheme 1 or scheme 3. No matter how
ber of iterations is larger. In Fig. 6, it is observed that, when the high the entrainment speed is, generally, schemes 2 and 4 appear
rolling speed is equal to or greater than U ¼ 0.1 m/s, the computa- to have similar performance but scheme 4 may be slightly better
tional speeds by schemes 2 and 4 are remarkably higher than those most of the time. Scheme 3 for jumping from K ¼ II directly to
by schemes 1 and 3. When the rolling speed is reduced down to K ¼ V is not preferred for the cases with smooth surfaces. Based

021502-8 / Vol. 138, APRIL 2016 Transactions of the ASME

Downloaded From: http://tribology.asmedigitalcollection.asme.org/ on 05/06/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use


Fig. 7 Continued

on the comparison above, scheme 4 is chosen to be the optimized average film thickness, i.e., HII > HIII > HIV > HV. When rolling
one for all the cases analyzed with smooth surfaces, unless other- speed is reduced down to 0.01 m/s, the converged average film
wise noted. thickness at each mesh level varies irregularly probably due to
Figure 7 shows the solution processes for the rough surface very weak hydrodynamic action, as shown in Fig. 7(c). Besides, it
cases with different schemes at U ¼ 1 m/s and U ¼ 0.01 m/s, is also observed that the higher the mesh density, the lower the
respectively. It is observed from Fig. 7(a) that, when rolling speed local fluctuations of average film thickness. Note that it can be
is U ¼ 1 m/s, the higher the mesh density, the lower the converged seen from Figs. 7(b) and 7(d) that, no matter which compound

Journal of Tribology APRIL 2016, Vol. 138 / 021502-9

Downloaded From: http://tribology.asmedigitalcollection.asme.org/ on 05/06/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use


6 Effect of Different Initial Values
Effect of different initial conditions has also been investigated
in the present study in order to examine the stability and reliability
of the solution approach with the PMD. In this section, the rolling
speed is U ¼ 1 m/s, and the maximum Hertzian pressure
Ph ¼ 1.86 GPa. The composite rms roughness is r ¼ 0.5 lm.
Obtained results for the additional rough surface EHL cases ana-
lyzed are summarized in Fig. 9, in which the red lines are the
results based on the initial value of pressure distribution from the
corresponding smooth surface solution under the same operating
conditions, and the blue lines are those from the Hertzian dry con-
tact solution. It appears that the effect of initial value on the tran-
sient mixed EHL solution becomes insignificant, and both the red
and blue lines get very close to each other after a sufficient num-
ber of time steps (e.g., 340–360 steps in this case). This indicates
that the final solutions are independent of the mesh arrangement
schemes used, and the numerical approach with each of the PMD
schemes is stable and reliable for this transient mixed EHL prob-
lem. It can also be seen that the initial value from the correspond-
Fig. 8 Comparison of computational speeds with different ing smooth surface solution appears to provide a more stable
schemes for cases with roughness performance in the solution process than the one from the Hertz-
ian dry contact theory.
mode is taken, after a sufficient number of iterations all the solu-
tions at the same speed become identical and globally stabilized 7 Conclusions
on mesh level 5, but small fluctuations still exist due to moving
surface irregularities. This indicates that the PMD method is sta- A PMD method for solving the mixed EHL problems was ini-
ble regardless of the compound mode used. In addition, it appears tially presented with a few preliminary sample solutions by Zhu
that with the rough surfaces scheme 3 may have a slightly better [20]. It has now been further explored in the present study based
performance than the others at U ¼ 1 m/s and 0.01 m/s, but the dif- on the simulation results obtained from a good number of cases
ferences among different compound modes are not large. with and without machined surface roughness in a wide range of
The comparison of computational speeds with different operating conditions. The conclusions drawn from the present
schemes for the cases with the rough surfaces is given in Fig. 8. It study include:
is found that, when the rolling speed is high, e.g., 10 m/s, the com- (1) The PMD solution procedure based on the semisystem
putational speed with scheme 4 is the highest. If the speed is low approach originally presented by Ai [11] and improved by
or moderate, the PMD approach with scheme 1 and scheme 3 Zhu and Hu [14,15,20] and others is capable of solving the
appears to be slightly better. Therefore, scheme 3 can be used mixed EHL problems effectively, and the solution conver-
except at high speeds. gence is ensured even with real-machined surface

Fig. 9 Effect of initial value on the transient mixed EHL solution

021502-10 / Vol. 138, APRIL 2016 Transactions of the ASME

Downloaded From: http://tribology.asmedigitalcollection.asme.org/ on 05/06/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use


roughness under severe operating conditions. Both the X, Y ¼ dimensionless coordinates, X ¼ x/a, Y ¼ y/b
hydrodynamic lubrication and the contact in mixed EHL a¼ pressure–viscosity exponent
can be handled with a unified model and numerical DT ¼ dimensionless time step length, DT ¼ UDt/a
approach. DX, DY ¼ dimensionless mesh sizes, DX ¼ Dx/a, DY ¼ Dy/b
(2) It has been demonstrated that with the PMD procedure, the g, go ¼ viscosity, and viscosity under ambient condition
solution process can be greatly accelerated in a wide range q, qo ¼ density of lubricant, and density under ambient
of operating conditions, in comparison with the solution condition
process with a single fixed discretization mesh, no matter
with or without surface roughness involved.
(3) Different solution processes with and without the PMD References
yield the same results after a sufficiently large number of [1] Ranger, A. P., Ettles, C. M. M., and Cameron, A., 1975, “The Solution of the
time steps. Also, different initial values and/or different Point Contact Elastohydrodynamic Problem,” Proc. R. Soc., London A, 346,
pp. 227–244.
compound modes in the PMD procedure lead to the same [2] Hamrock, B. J., and Dowson, D., 1976, “Isothermal Elastohydrodynamic Lubri-
final solution at the selected highest mesh level. This indi- cation of Point Contacts—Part 1: Theoretical Formulation,” ASME J. Lubr.
cates that the numerical approaches are stable and reliable Technol., 98(2), pp. 223–229.
for transient mixed EHL problems. [3] Hamrock, B. J., and Dowson, D., 1976, “Isothermal Elastohydrodynamic Lubri-
cation of Point Contacts—Part 2: Ellipticity Parameter Results,” ASME J.
(4) The effect of different compound modes on the EHL solu- Lubr. Technol., 98(3), pp. 375–383.
tion process is systematically investigated in a wide range [4] Hamrock, B. J., and Dowson, D., 1977, “Isothermal Elastohydrodynamic Lubri-
of operating conditions in order to optimize the PMD pro- cation of Point Contacts—Part 3: Fully Flooded Results,” ASME J. Lubr. Tech-
cedure. It has been found that, for smooth surface cases, nol., 99(2), pp. 264–276.
[5] Evans, H. P., and Snidle, R. W., 1981, “Inverse Solution of Reynolds Equation
scheme 4 may generally be the best choice except that at of Lubrication Under Point Contact Elastohydrodynamic Conditions,” ASME J.
very low speeds scheme 1 may be slightly better. For rough Lubr. Technol., 103(4), pp. 539–546.
surface cases, however, it is observed that, when the rolling [6] Evans, H. P., and Snidle, R. W., 1982, “The Elastohydrodynamic Lubrication
speed is high, scheme 4 may offer the highest efficiency, of Point Contacts at Heavy Loads,” Proc. R. Soc. A, 382(1782), pp. 183–199.
[7] Zhu, D., and Wen, S. Z., 1984, “A Full Numerical Solution for the
but if the speed is low or moderate, scheme 3 appears to be Thermoelasto-Hydrodynamic Problem in Elliptical Contacts,” ASME J. Tribol.,
generally more appropriate. 106(2), pp. 246–254.
[8] Lubrecht, A. A., 1987, “The Numerical Solution of Elastohydrodynamic Lubri-
cated Line and Point Contact Problems Using Multigrid Techniques,” Ph.D.
Acknowledgment thesis, University of Twente, Enschede, The Netherlands.
[9] Venner, C. H., 1991, “Multilevel Solution of EHL Line and Point Contact Prob-
This study was partially supported by the National Science Foun- lems,” Ph.D. thesis, University of Twente, Enschede, The Netherlands.
dation of China (NSFC) Project Nos. 51435001 and 51375506. Wei [10] Kim, K. H., and Sadeghi, F., 1992, “Three Dimensional Temperature Distribu-
tion in EHD Lubrication—Part I: Circular Contact,” ASME J. Tribol., 114(1),
Pu would also like to acknowledge the Research Foundation from pp. 32–41.
the Sichuan University (No. SCUMIAOZI2014-1-1). [11] Ai, X. L., 1993, “Numerical Analyses of Elastohydrodynamically Lubricated
Line and Point Contacts With Rough Surfaces By Using Semi-System and Mul-
tigrid Methods,” Ph. D. thesis, Northwestern University, Evanston, IL.
Nomenclature [12] Zhu, D., and Ai, X. L., 1997, “Point Contact EHL Based on Optically Measured
3-Dimensional Rough Surfaces,” ASME J. Tribol., 119(3), pp. 375–384.
a ¼ semi-axis of the Hertzian contact ellipse in the x- [13] Venner, C. H., and Lubrecht, A. A., 2000, “Multilevel Methods in Lubrication,”
direction Elsevier Tribology Series, Vol. 37, Elsevier, Amsterdam.
b ¼ semi-axis of the Hertzian contact ellipse in the y- [14] Zhu, D., and Hu, Y. Z., 1999, “The Study of Transition From Full Film Elasto-
direction hydrodynamic to Mixed and Boundary Lubrication,” The Advancing Frontier
of Engineering Tribology, 1999 STLE/ASME H. S. Cheng Tribology Surveil-
E0 ¼ effective Young’s modulus lance, STLE, Park Ridge, IL, pp. 150–156.
h ¼ local film thickness (or gap) [15] Hu, Y. Z., and Zhu, D., 2000, “A Full Numerical Solution to the Mixed Lubri-
H ¼ h/a, dimensionless film thickness cation in Point Contacts,” ASME J. Tribol., 122(1), pp. 1–9.
ha ¼ average film thicknesses evaluated in the central part of [16] Morales-Espejel, G. E., Dumont, M. L., Lugt, P. M., and Olver, A. V., 2005, “A
Limiting Solution for the Dependence of Film Thickness on Velocity in EHL
the Hertzian contact zone Contacts With Very Thin Films,” Tribol. Trans., 48(3), pp. 317–324.
hc, hm ¼ central and minimum film thicknesses, respectively [17] Zhu, D., 2005, “Written Discussion on: A Limiting Solution for the Depend-
HK ¼ converged dimensionless central film thickness on ence of Film Thickness on Velocity in EHL Contacts With Very Thin Films,”
mesh level K Tribol. Trans., 48(3), pp. 324–326.
[18] Venner, C. H., 2005, “EHL Film Thickness Computations at Low Speeds: Risk
H1 ¼ dimensionless film thickness theoretically from an of Artificial Trends as a Result of Poor Accuracy and Implications for Mixed
infinitely dense mesh Lubrication Modeling,” Proc. Inst. Mech. Eng., Part J, 219(4), pp. 285–290.
K ¼ mesh level number [19] Liu, Y. C., Wang, Q., Wang, W. Z., Hu, Y. Z., and Zhu, D., 2006, “Effects of
p ¼ pressure, P ¼ p/Ph, dimensionless pressure Differential Scheme and Mesh Density on EHL Film Thickness in Point Con-
tacts,” ASME J. Tribol., 128(3), pp. 641–653.
Ph ¼ nominal maximum Hertzian contact pressure [20] Zhu, D., 2007, “On Some Aspects in Numerical Solution of Thin-Film and
Rx ¼ effective radius of curvature in the x-direction Mixed EHL,” Proc. Inst. Mech. Eng., Part J, 221(5), pp. 561–579.
Ry ¼ effective radius of curvature in the y-direction [21] Liu, S. B., Wang, Q., and Liu, G., 2000, “A Versatile Method of Discrete Con-
t ¼ time, T ¼ tU/a, dimensionless time volution and FFT (DC-FFT) for Contact Analyses,” Wear, 243(1–2), pp.
101–111.
Td ¼ computation time with the single fixed computational [22] Wang, W. Z., Wang, H., Liu, Y. C., Hu, Y. Z., and Zhu, D., 2003, “A Compara-
mesh tive Study of the Methods for Calculation of Surface Elastic Deformation,”
Ti ¼ computing time required by scheme i (i ¼ 1, 2, 3, 4) Proc. Inst. Mech. Eng., Part J, 217(2), pp. 145–153.
U ¼ (u1 þ u2)/2, rolling velocity in the x-direction [23] Zhu, D., Liu, Y., and Wang, Q., 2014, “On the Numerical Accuracy of Rough
Surface EHL Solution,” Tribol. Trans., 57(4), pp. 570–580.
u1, u2 ¼ velocities of surface 1 and surface 2 in the x-direction, [24] Lubrecht, A. A., and Venner, C. H., 1999, “Elastohydrodynamic Lubrication of
respectively Rough Surfaces,” Proc. Inst. Mech. Eng., Part J, 213(5), pp. 397–404.
V ¼ surface elastic deformation [25] Mourier, L., Mazuyer, D., Lubrecht, A. A., and Donnet, C., 2006, “Transient
W ¼ applied load Increase of Film Thickness in Micro-Textured EHL Contacts,” Tribol. Int.,
39(12), pp. 1745–1756.
Wc ¼ contact load ratio, defined as the load supported by [26] Zhu, D., and Hu, Y. Z., 2001, “Effects of Rough Surface Topography and Ori-
asperity contacts divided by the total load entation on the Characteristics of EHD and Mixed Lubrication in Both Circular
x, y ¼ coordinates and Elliptical Contacts,” Tribol. Trans., 44(3), pp. 391–398.

Journal of Tribology APRIL 2016, Vol. 138 / 021502-11

Downloaded From: http://tribology.asmedigitalcollection.asme.org/ on 05/06/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

You might also like